| { | |
| "best_global_step": 3000, | |
| "best_metric": 3.711540102433069, | |
| "best_model_checkpoint": "checkpoints/gpt-2_seq1024_mla192-96-192/checkpoint-3000", | |
| "epoch": 0.9042272624519629, | |
| "eval_steps": 300, | |
| "global_step": 3000, | |
| "is_hyper_param_search": false, | |
| "is_local_process_zero": true, | |
| "is_world_process_zero": true, | |
| "log_history": [ | |
| { | |
| "epoch": 0.006028181749679753, | |
| "grad_norm": 2.0335798263549805, | |
| "learning_rate": 3.166666666666667e-05, | |
| "loss": 10.3669, | |
| "step": 20 | |
| }, | |
| { | |
| "epoch": 0.012056363499359506, | |
| "grad_norm": 1.616795539855957, | |
| "learning_rate": 6.500000000000001e-05, | |
| "loss": 9.2122, | |
| "step": 40 | |
| }, | |
| { | |
| "epoch": 0.01808454524903926, | |
| "grad_norm": 1.1775591373443604, | |
| "learning_rate": 9.833333333333333e-05, | |
| "loss": 8.2448, | |
| "step": 60 | |
| }, | |
| { | |
| "epoch": 0.02411272699871901, | |
| "grad_norm": 0.9382872581481934, | |
| "learning_rate": 0.00013166666666666665, | |
| "loss": 7.6529, | |
| "step": 80 | |
| }, | |
| { | |
| "epoch": 0.030140908748398764, | |
| "grad_norm": 0.9344885349273682, | |
| "learning_rate": 0.000165, | |
| "loss": 7.3608, | |
| "step": 100 | |
| }, | |
| { | |
| "epoch": 0.03616909049807852, | |
| "grad_norm": 2.1915502548217773, | |
| "learning_rate": 0.00019833333333333335, | |
| "loss": 7.0825, | |
| "step": 120 | |
| }, | |
| { | |
| "epoch": 0.04219727224775827, | |
| "grad_norm": 2.2511191368103027, | |
| "learning_rate": 0.00023166666666666667, | |
| "loss": 6.8945, | |
| "step": 140 | |
| }, | |
| { | |
| "epoch": 0.04822545399743802, | |
| "grad_norm": 1.1455268859863281, | |
| "learning_rate": 0.00026500000000000004, | |
| "loss": 6.7276, | |
| "step": 160 | |
| }, | |
| { | |
| "epoch": 0.05425363574711778, | |
| "grad_norm": 0.925327718257904, | |
| "learning_rate": 0.00029833333333333334, | |
| "loss": 6.5657, | |
| "step": 180 | |
| }, | |
| { | |
| "epoch": 0.06028181749679753, | |
| "grad_norm": 1.1821664571762085, | |
| "learning_rate": 0.0003316666666666667, | |
| "loss": 6.4371, | |
| "step": 200 | |
| }, | |
| { | |
| "epoch": 0.06630999924647728, | |
| "grad_norm": 1.0635254383087158, | |
| "learning_rate": 0.000365, | |
| "loss": 6.2975, | |
| "step": 220 | |
| }, | |
| { | |
| "epoch": 0.07233818099615703, | |
| "grad_norm": 0.953662097454071, | |
| "learning_rate": 0.00039833333333333333, | |
| "loss": 6.1737, | |
| "step": 240 | |
| }, | |
| { | |
| "epoch": 0.07836636274583679, | |
| "grad_norm": 1.062455415725708, | |
| "learning_rate": 0.0004316666666666667, | |
| "loss": 6.0755, | |
| "step": 260 | |
| }, | |
| { | |
| "epoch": 0.08439454449551655, | |
| "grad_norm": 1.1244603395462036, | |
| "learning_rate": 0.000465, | |
| "loss": 5.941, | |
| "step": 280 | |
| }, | |
| { | |
| "epoch": 0.09042272624519629, | |
| "grad_norm": 1.1435415744781494, | |
| "learning_rate": 0.0004983333333333334, | |
| "loss": 5.8408, | |
| "step": 300 | |
| }, | |
| { | |
| "epoch": 0.09042272624519629, | |
| "eval_loss": 5.775358345603854, | |
| "eval_perplexity": 322.2598944464712, | |
| "eval_runtime": 112.8805, | |
| "eval_samples_per_second": 14.697, | |
| "eval_steps_per_second": 0.23, | |
| "step": 300 | |
| }, | |
| { | |
| "epoch": 0.09645090799487605, | |
| "grad_norm": 0.6207288503646851, | |
| "learning_rate": 0.0004964814814814814, | |
| "loss": 5.7142, | |
| "step": 320 | |
| }, | |
| { | |
| "epoch": 0.1024790897445558, | |
| "grad_norm": 0.7662326693534851, | |
| "learning_rate": 0.0004927777777777777, | |
| "loss": 5.6204, | |
| "step": 340 | |
| }, | |
| { | |
| "epoch": 0.10850727149423556, | |
| "grad_norm": 0.9737507104873657, | |
| "learning_rate": 0.0004890740740740741, | |
| "loss": 5.5194, | |
| "step": 360 | |
| }, | |
| { | |
| "epoch": 0.1145354532439153, | |
| "grad_norm": 0.8488203883171082, | |
| "learning_rate": 0.0004853703703703704, | |
| "loss": 5.4269, | |
| "step": 380 | |
| }, | |
| { | |
| "epoch": 0.12056363499359506, | |
| "grad_norm": 0.6605532765388489, | |
| "learning_rate": 0.0004816666666666667, | |
| "loss": 5.3382, | |
| "step": 400 | |
| }, | |
| { | |
| "epoch": 0.1265918167432748, | |
| "grad_norm": 0.8578091859817505, | |
| "learning_rate": 0.00047796296296296297, | |
| "loss": 5.2569, | |
| "step": 420 | |
| }, | |
| { | |
| "epoch": 0.13261999849295456, | |
| "grad_norm": 0.6136470437049866, | |
| "learning_rate": 0.0004742592592592593, | |
| "loss": 5.2025, | |
| "step": 440 | |
| }, | |
| { | |
| "epoch": 0.1386481802426343, | |
| "grad_norm": 0.9338096976280212, | |
| "learning_rate": 0.00047055555555555555, | |
| "loss": 5.1277, | |
| "step": 460 | |
| }, | |
| { | |
| "epoch": 0.14467636199231407, | |
| "grad_norm": 0.6009561419487, | |
| "learning_rate": 0.00046685185185185187, | |
| "loss": 5.0709, | |
| "step": 480 | |
| }, | |
| { | |
| "epoch": 0.15070454374199382, | |
| "grad_norm": 0.8465147614479065, | |
| "learning_rate": 0.00046314814814814813, | |
| "loss": 5.0067, | |
| "step": 500 | |
| }, | |
| { | |
| "epoch": 0.15673272549167358, | |
| "grad_norm": 0.7476010918617249, | |
| "learning_rate": 0.00045944444444444445, | |
| "loss": 4.9605, | |
| "step": 520 | |
| }, | |
| { | |
| "epoch": 0.16276090724135334, | |
| "grad_norm": 0.9261116981506348, | |
| "learning_rate": 0.0004557407407407407, | |
| "loss": 4.9077, | |
| "step": 540 | |
| }, | |
| { | |
| "epoch": 0.1687890889910331, | |
| "grad_norm": 0.6120880842208862, | |
| "learning_rate": 0.00045203703703703703, | |
| "loss": 4.8659, | |
| "step": 560 | |
| }, | |
| { | |
| "epoch": 0.17481727074071282, | |
| "grad_norm": 0.9202558398246765, | |
| "learning_rate": 0.0004483333333333333, | |
| "loss": 4.8183, | |
| "step": 580 | |
| }, | |
| { | |
| "epoch": 0.18084545249039258, | |
| "grad_norm": 0.5459677577018738, | |
| "learning_rate": 0.00044462962962962967, | |
| "loss": 4.7835, | |
| "step": 600 | |
| }, | |
| { | |
| "epoch": 0.18084545249039258, | |
| "eval_loss": 4.719039675616339, | |
| "eval_perplexity": 112.06058646521387, | |
| "eval_runtime": 112.3965, | |
| "eval_samples_per_second": 14.76, | |
| "eval_steps_per_second": 0.231, | |
| "step": 600 | |
| }, | |
| { | |
| "epoch": 0.18687363424007233, | |
| "grad_norm": 0.8926022052764893, | |
| "learning_rate": 0.00044092592592592594, | |
| "loss": 4.7306, | |
| "step": 620 | |
| }, | |
| { | |
| "epoch": 0.1929018159897521, | |
| "grad_norm": 0.7751118540763855, | |
| "learning_rate": 0.00043722222222222225, | |
| "loss": 4.7086, | |
| "step": 640 | |
| }, | |
| { | |
| "epoch": 0.19892999773943185, | |
| "grad_norm": 0.764580488204956, | |
| "learning_rate": 0.0004335185185185185, | |
| "loss": 4.668, | |
| "step": 660 | |
| }, | |
| { | |
| "epoch": 0.2049581794891116, | |
| "grad_norm": 0.8693467974662781, | |
| "learning_rate": 0.00042981481481481484, | |
| "loss": 4.635, | |
| "step": 680 | |
| }, | |
| { | |
| "epoch": 0.21098636123879136, | |
| "grad_norm": 0.5976945161819458, | |
| "learning_rate": 0.0004261111111111111, | |
| "loss": 4.6075, | |
| "step": 700 | |
| }, | |
| { | |
| "epoch": 0.21701454298847112, | |
| "grad_norm": 0.4756850302219391, | |
| "learning_rate": 0.0004224074074074074, | |
| "loss": 4.5614, | |
| "step": 720 | |
| }, | |
| { | |
| "epoch": 0.22304272473815084, | |
| "grad_norm": 0.8641749024391174, | |
| "learning_rate": 0.0004187037037037037, | |
| "loss": 4.5238, | |
| "step": 740 | |
| }, | |
| { | |
| "epoch": 0.2290709064878306, | |
| "grad_norm": 0.7556053996086121, | |
| "learning_rate": 0.000415, | |
| "loss": 4.4955, | |
| "step": 760 | |
| }, | |
| { | |
| "epoch": 0.23509908823751036, | |
| "grad_norm": 0.663978099822998, | |
| "learning_rate": 0.00041129629629629627, | |
| "loss": 4.465, | |
| "step": 780 | |
| }, | |
| { | |
| "epoch": 0.2411272699871901, | |
| "grad_norm": 0.8250914812088013, | |
| "learning_rate": 0.00040759259259259264, | |
| "loss": 4.4335, | |
| "step": 800 | |
| }, | |
| { | |
| "epoch": 0.24715545173686987, | |
| "grad_norm": 0.9232884645462036, | |
| "learning_rate": 0.0004038888888888889, | |
| "loss": 4.3991, | |
| "step": 820 | |
| }, | |
| { | |
| "epoch": 0.2531836334865496, | |
| "grad_norm": 0.7927812337875366, | |
| "learning_rate": 0.0004001851851851852, | |
| "loss": 4.3797, | |
| "step": 840 | |
| }, | |
| { | |
| "epoch": 0.2592118152362294, | |
| "grad_norm": 0.6007223129272461, | |
| "learning_rate": 0.0003964814814814815, | |
| "loss": 4.3527, | |
| "step": 860 | |
| }, | |
| { | |
| "epoch": 0.2652399969859091, | |
| "grad_norm": 0.9147293567657471, | |
| "learning_rate": 0.0003927777777777778, | |
| "loss": 4.3308, | |
| "step": 880 | |
| }, | |
| { | |
| "epoch": 0.2712681787355889, | |
| "grad_norm": 0.7305644750595093, | |
| "learning_rate": 0.00038907407407407407, | |
| "loss": 4.3067, | |
| "step": 900 | |
| }, | |
| { | |
| "epoch": 0.2712681787355889, | |
| "eval_loss": 4.257442998791508, | |
| "eval_perplexity": 70.62915353070326, | |
| "eval_runtime": 112.5319, | |
| "eval_samples_per_second": 14.742, | |
| "eval_steps_per_second": 0.231, | |
| "step": 900 | |
| }, | |
| { | |
| "epoch": 0.2772963604852686, | |
| "grad_norm": 0.6429543495178223, | |
| "learning_rate": 0.0003853703703703704, | |
| "loss": 4.2872, | |
| "step": 920 | |
| }, | |
| { | |
| "epoch": 0.2833245422349484, | |
| "grad_norm": 0.8753812313079834, | |
| "learning_rate": 0.00038166666666666666, | |
| "loss": 4.2743, | |
| "step": 940 | |
| }, | |
| { | |
| "epoch": 0.28935272398462814, | |
| "grad_norm": 0.7068883180618286, | |
| "learning_rate": 0.000377962962962963, | |
| "loss": 4.2521, | |
| "step": 960 | |
| }, | |
| { | |
| "epoch": 0.29538090573430786, | |
| "grad_norm": 0.725367546081543, | |
| "learning_rate": 0.00037425925925925924, | |
| "loss": 4.2405, | |
| "step": 980 | |
| }, | |
| { | |
| "epoch": 0.30140908748398765, | |
| "grad_norm": 0.680828869342804, | |
| "learning_rate": 0.0003705555555555556, | |
| "loss": 4.2247, | |
| "step": 1000 | |
| }, | |
| { | |
| "epoch": 0.3074372692336674, | |
| "grad_norm": 0.7420363426208496, | |
| "learning_rate": 0.0003668518518518519, | |
| "loss": 4.2071, | |
| "step": 1020 | |
| }, | |
| { | |
| "epoch": 0.31346545098334716, | |
| "grad_norm": 0.978520929813385, | |
| "learning_rate": 0.0003631481481481482, | |
| "loss": 4.1864, | |
| "step": 1040 | |
| }, | |
| { | |
| "epoch": 0.3194936327330269, | |
| "grad_norm": 0.684148371219635, | |
| "learning_rate": 0.00035944444444444446, | |
| "loss": 4.1818, | |
| "step": 1060 | |
| }, | |
| { | |
| "epoch": 0.3255218144827067, | |
| "grad_norm": 0.6693135499954224, | |
| "learning_rate": 0.0003557407407407408, | |
| "loss": 4.1668, | |
| "step": 1080 | |
| }, | |
| { | |
| "epoch": 0.3315499962323864, | |
| "grad_norm": 0.6347187757492065, | |
| "learning_rate": 0.00035203703703703704, | |
| "loss": 4.1489, | |
| "step": 1100 | |
| }, | |
| { | |
| "epoch": 0.3375781779820662, | |
| "grad_norm": 0.6263527870178223, | |
| "learning_rate": 0.00034833333333333336, | |
| "loss": 4.1432, | |
| "step": 1120 | |
| }, | |
| { | |
| "epoch": 0.3436063597317459, | |
| "grad_norm": 0.6609379053115845, | |
| "learning_rate": 0.0003446296296296296, | |
| "loss": 4.1266, | |
| "step": 1140 | |
| }, | |
| { | |
| "epoch": 0.34963454148142564, | |
| "grad_norm": 0.846114456653595, | |
| "learning_rate": 0.0003409259259259259, | |
| "loss": 4.1157, | |
| "step": 1160 | |
| }, | |
| { | |
| "epoch": 0.3556627232311054, | |
| "grad_norm": 1.0905325412750244, | |
| "learning_rate": 0.0003372222222222222, | |
| "loss": 4.102, | |
| "step": 1180 | |
| }, | |
| { | |
| "epoch": 0.36169090498078516, | |
| "grad_norm": 0.6602559685707092, | |
| "learning_rate": 0.0003335185185185185, | |
| "loss": 4.1018, | |
| "step": 1200 | |
| }, | |
| { | |
| "epoch": 0.36169090498078516, | |
| "eval_loss": 4.045143906471234, | |
| "eval_perplexity": 57.11940529478326, | |
| "eval_runtime": 112.5597, | |
| "eval_samples_per_second": 14.739, | |
| "eval_steps_per_second": 0.231, | |
| "step": 1200 | |
| }, | |
| { | |
| "epoch": 0.36771908673046494, | |
| "grad_norm": 0.6741213798522949, | |
| "learning_rate": 0.00032981481481481485, | |
| "loss": 4.0838, | |
| "step": 1220 | |
| }, | |
| { | |
| "epoch": 0.37374726848014467, | |
| "grad_norm": 0.8439180850982666, | |
| "learning_rate": 0.0003261111111111111, | |
| "loss": 4.0786, | |
| "step": 1240 | |
| }, | |
| { | |
| "epoch": 0.37977545022982445, | |
| "grad_norm": 0.6376078724861145, | |
| "learning_rate": 0.00032240740740740743, | |
| "loss": 4.0629, | |
| "step": 1260 | |
| }, | |
| { | |
| "epoch": 0.3858036319795042, | |
| "grad_norm": 0.8104387521743774, | |
| "learning_rate": 0.0003187037037037037, | |
| "loss": 4.0562, | |
| "step": 1280 | |
| }, | |
| { | |
| "epoch": 0.3918318137291839, | |
| "grad_norm": 0.74709552526474, | |
| "learning_rate": 0.000315, | |
| "loss": 4.0475, | |
| "step": 1300 | |
| }, | |
| { | |
| "epoch": 0.3978599954788637, | |
| "grad_norm": 0.7379426956176758, | |
| "learning_rate": 0.0003112962962962963, | |
| "loss": 4.0371, | |
| "step": 1320 | |
| }, | |
| { | |
| "epoch": 0.4038881772285434, | |
| "grad_norm": 0.9285239577293396, | |
| "learning_rate": 0.0003075925925925926, | |
| "loss": 4.0355, | |
| "step": 1340 | |
| }, | |
| { | |
| "epoch": 0.4099163589782232, | |
| "grad_norm": 0.6420581936836243, | |
| "learning_rate": 0.00030388888888888886, | |
| "loss": 4.0266, | |
| "step": 1360 | |
| }, | |
| { | |
| "epoch": 0.41594454072790293, | |
| "grad_norm": 0.6401010751724243, | |
| "learning_rate": 0.0003001851851851852, | |
| "loss": 4.0159, | |
| "step": 1380 | |
| }, | |
| { | |
| "epoch": 0.4219727224775827, | |
| "grad_norm": 0.5986769795417786, | |
| "learning_rate": 0.00029648148148148144, | |
| "loss": 4.0089, | |
| "step": 1400 | |
| }, | |
| { | |
| "epoch": 0.42800090422726245, | |
| "grad_norm": 0.7083526253700256, | |
| "learning_rate": 0.0002927777777777778, | |
| "loss": 4.0079, | |
| "step": 1420 | |
| }, | |
| { | |
| "epoch": 0.43402908597694223, | |
| "grad_norm": 0.613379955291748, | |
| "learning_rate": 0.0002890740740740741, | |
| "loss": 3.9933, | |
| "step": 1440 | |
| }, | |
| { | |
| "epoch": 0.44005726772662196, | |
| "grad_norm": 0.5723183155059814, | |
| "learning_rate": 0.0002853703703703704, | |
| "loss": 3.9961, | |
| "step": 1460 | |
| }, | |
| { | |
| "epoch": 0.4460854494763017, | |
| "grad_norm": 0.6976141929626465, | |
| "learning_rate": 0.00028166666666666666, | |
| "loss": 3.9877, | |
| "step": 1480 | |
| }, | |
| { | |
| "epoch": 0.45211363122598147, | |
| "grad_norm": 0.6135937571525574, | |
| "learning_rate": 0.000277962962962963, | |
| "loss": 3.9795, | |
| "step": 1500 | |
| }, | |
| { | |
| "epoch": 0.45211363122598147, | |
| "eval_loss": 3.9260568372578377, | |
| "eval_perplexity": 50.70663841626086, | |
| "eval_runtime": 112.6508, | |
| "eval_samples_per_second": 14.727, | |
| "eval_steps_per_second": 0.231, | |
| "step": 1500 | |
| }, | |
| { | |
| "epoch": 0.4581418129756612, | |
| "grad_norm": 0.8446961641311646, | |
| "learning_rate": 0.00027425925925925925, | |
| "loss": 3.9717, | |
| "step": 1520 | |
| }, | |
| { | |
| "epoch": 0.464169994725341, | |
| "grad_norm": 0.5484296679496765, | |
| "learning_rate": 0.00027055555555555557, | |
| "loss": 3.9681, | |
| "step": 1540 | |
| }, | |
| { | |
| "epoch": 0.4701981764750207, | |
| "grad_norm": 0.6464419364929199, | |
| "learning_rate": 0.00026685185185185183, | |
| "loss": 3.9544, | |
| "step": 1560 | |
| }, | |
| { | |
| "epoch": 0.4762263582247005, | |
| "grad_norm": 0.6416765451431274, | |
| "learning_rate": 0.00026314814814814815, | |
| "loss": 3.9548, | |
| "step": 1580 | |
| }, | |
| { | |
| "epoch": 0.4822545399743802, | |
| "grad_norm": 0.5626780390739441, | |
| "learning_rate": 0.0002594444444444444, | |
| "loss": 3.9384, | |
| "step": 1600 | |
| }, | |
| { | |
| "epoch": 0.48828272172405995, | |
| "grad_norm": 0.9356988668441772, | |
| "learning_rate": 0.0002557407407407408, | |
| "loss": 3.946, | |
| "step": 1620 | |
| }, | |
| { | |
| "epoch": 0.49431090347373974, | |
| "grad_norm": 0.8774187564849854, | |
| "learning_rate": 0.00025203703703703705, | |
| "loss": 3.9363, | |
| "step": 1640 | |
| }, | |
| { | |
| "epoch": 0.5003390852234195, | |
| "grad_norm": 0.5745379328727722, | |
| "learning_rate": 0.0002483333333333333, | |
| "loss": 3.9333, | |
| "step": 1660 | |
| }, | |
| { | |
| "epoch": 0.5063672669730992, | |
| "grad_norm": 0.6884981393814087, | |
| "learning_rate": 0.00024462962962962963, | |
| "loss": 3.929, | |
| "step": 1680 | |
| }, | |
| { | |
| "epoch": 0.512395448722779, | |
| "grad_norm": 0.7028161287307739, | |
| "learning_rate": 0.00024092592592592593, | |
| "loss": 3.9148, | |
| "step": 1700 | |
| }, | |
| { | |
| "epoch": 0.5184236304724588, | |
| "grad_norm": 0.571987509727478, | |
| "learning_rate": 0.00023722222222222222, | |
| "loss": 3.9146, | |
| "step": 1720 | |
| }, | |
| { | |
| "epoch": 0.5244518122221385, | |
| "grad_norm": 0.5669906735420227, | |
| "learning_rate": 0.0002335185185185185, | |
| "loss": 3.9141, | |
| "step": 1740 | |
| }, | |
| { | |
| "epoch": 0.5304799939718182, | |
| "grad_norm": 0.6806395053863525, | |
| "learning_rate": 0.0002298148148148148, | |
| "loss": 3.9123, | |
| "step": 1760 | |
| }, | |
| { | |
| "epoch": 0.536508175721498, | |
| "grad_norm": 0.610780656337738, | |
| "learning_rate": 0.00022611111111111112, | |
| "loss": 3.8998, | |
| "step": 1780 | |
| }, | |
| { | |
| "epoch": 0.5425363574711778, | |
| "grad_norm": 0.6254203915596008, | |
| "learning_rate": 0.0002224074074074074, | |
| "loss": 3.9015, | |
| "step": 1800 | |
| }, | |
| { | |
| "epoch": 0.5425363574711778, | |
| "eval_loss": 3.8490570093839285, | |
| "eval_perplexity": 46.94877010121682, | |
| "eval_runtime": 112.5759, | |
| "eval_samples_per_second": 14.737, | |
| "eval_steps_per_second": 0.231, | |
| "step": 1800 | |
| }, | |
| { | |
| "epoch": 0.5485645392208575, | |
| "grad_norm": 0.7416952848434448, | |
| "learning_rate": 0.0002187037037037037, | |
| "loss": 3.8982, | |
| "step": 1820 | |
| }, | |
| { | |
| "epoch": 0.5545927209705372, | |
| "grad_norm": 0.6967796087265015, | |
| "learning_rate": 0.000215, | |
| "loss": 3.8907, | |
| "step": 1840 | |
| }, | |
| { | |
| "epoch": 0.560620902720217, | |
| "grad_norm": 0.6138114333152771, | |
| "learning_rate": 0.00021129629629629629, | |
| "loss": 3.8815, | |
| "step": 1860 | |
| }, | |
| { | |
| "epoch": 0.5666490844698968, | |
| "grad_norm": 0.6929096579551697, | |
| "learning_rate": 0.0002075925925925926, | |
| "loss": 3.881, | |
| "step": 1880 | |
| }, | |
| { | |
| "epoch": 0.5726772662195765, | |
| "grad_norm": 0.658118724822998, | |
| "learning_rate": 0.0002038888888888889, | |
| "loss": 3.8726, | |
| "step": 1900 | |
| }, | |
| { | |
| "epoch": 0.5787054479692563, | |
| "grad_norm": 0.631330132484436, | |
| "learning_rate": 0.0002001851851851852, | |
| "loss": 3.876, | |
| "step": 1920 | |
| }, | |
| { | |
| "epoch": 0.584733629718936, | |
| "grad_norm": 0.47665688395500183, | |
| "learning_rate": 0.00019648148148148148, | |
| "loss": 3.8681, | |
| "step": 1940 | |
| }, | |
| { | |
| "epoch": 0.5907618114686157, | |
| "grad_norm": 0.5668326616287231, | |
| "learning_rate": 0.00019277777777777777, | |
| "loss": 3.8684, | |
| "step": 1960 | |
| }, | |
| { | |
| "epoch": 0.5967899932182955, | |
| "grad_norm": 0.5730007886886597, | |
| "learning_rate": 0.00018907407407407406, | |
| "loss": 3.858, | |
| "step": 1980 | |
| }, | |
| { | |
| "epoch": 0.6028181749679753, | |
| "grad_norm": 0.6014134287834167, | |
| "learning_rate": 0.00018537037037037038, | |
| "loss": 3.8586, | |
| "step": 2000 | |
| }, | |
| { | |
| "epoch": 0.6088463567176551, | |
| "grad_norm": 0.5977863073348999, | |
| "learning_rate": 0.00018166666666666667, | |
| "loss": 3.8539, | |
| "step": 2020 | |
| }, | |
| { | |
| "epoch": 0.6148745384673348, | |
| "grad_norm": 0.5913246870040894, | |
| "learning_rate": 0.00017796296296296296, | |
| "loss": 3.8485, | |
| "step": 2040 | |
| }, | |
| { | |
| "epoch": 0.6209027202170145, | |
| "grad_norm": 0.7566778063774109, | |
| "learning_rate": 0.00017425925925925926, | |
| "loss": 3.8486, | |
| "step": 2060 | |
| }, | |
| { | |
| "epoch": 0.6269309019666943, | |
| "grad_norm": 0.601656973361969, | |
| "learning_rate": 0.00017055555555555555, | |
| "loss": 3.8464, | |
| "step": 2080 | |
| }, | |
| { | |
| "epoch": 0.632959083716374, | |
| "grad_norm": 0.5155408382415771, | |
| "learning_rate": 0.00016685185185185187, | |
| "loss": 3.8412, | |
| "step": 2100 | |
| }, | |
| { | |
| "epoch": 0.632959083716374, | |
| "eval_loss": 3.7936583890736095, | |
| "eval_perplexity": 44.41860392831647, | |
| "eval_runtime": 112.4714, | |
| "eval_samples_per_second": 14.75, | |
| "eval_steps_per_second": 0.231, | |
| "step": 2100 | |
| }, | |
| { | |
| "epoch": 0.6389872654660538, | |
| "grad_norm": 0.5241331458091736, | |
| "learning_rate": 0.00016314814814814816, | |
| "loss": 3.8399, | |
| "step": 2120 | |
| }, | |
| { | |
| "epoch": 0.6450154472157336, | |
| "grad_norm": 0.5489856600761414, | |
| "learning_rate": 0.00015944444444444445, | |
| "loss": 3.8374, | |
| "step": 2140 | |
| }, | |
| { | |
| "epoch": 0.6510436289654133, | |
| "grad_norm": 0.45912450551986694, | |
| "learning_rate": 0.00015574074074074074, | |
| "loss": 3.8342, | |
| "step": 2160 | |
| }, | |
| { | |
| "epoch": 0.657071810715093, | |
| "grad_norm": 0.48803767561912537, | |
| "learning_rate": 0.00015203703703703703, | |
| "loss": 3.8333, | |
| "step": 2180 | |
| }, | |
| { | |
| "epoch": 0.6630999924647728, | |
| "grad_norm": 0.6220065355300903, | |
| "learning_rate": 0.00014833333333333335, | |
| "loss": 3.8292, | |
| "step": 2200 | |
| }, | |
| { | |
| "epoch": 0.6691281742144526, | |
| "grad_norm": 0.588616132736206, | |
| "learning_rate": 0.00014462962962962964, | |
| "loss": 3.8256, | |
| "step": 2220 | |
| }, | |
| { | |
| "epoch": 0.6751563559641324, | |
| "grad_norm": 0.5235931873321533, | |
| "learning_rate": 0.00014092592592592594, | |
| "loss": 3.8246, | |
| "step": 2240 | |
| }, | |
| { | |
| "epoch": 0.681184537713812, | |
| "grad_norm": 0.43203026056289673, | |
| "learning_rate": 0.00013722222222222223, | |
| "loss": 3.8113, | |
| "step": 2260 | |
| }, | |
| { | |
| "epoch": 0.6872127194634918, | |
| "grad_norm": 0.5042020082473755, | |
| "learning_rate": 0.00013351851851851852, | |
| "loss": 3.8178, | |
| "step": 2280 | |
| }, | |
| { | |
| "epoch": 0.6932409012131716, | |
| "grad_norm": 0.5330100059509277, | |
| "learning_rate": 0.00012981481481481484, | |
| "loss": 3.8115, | |
| "step": 2300 | |
| }, | |
| { | |
| "epoch": 0.6992690829628513, | |
| "grad_norm": 0.5101568698883057, | |
| "learning_rate": 0.00012611111111111113, | |
| "loss": 3.8047, | |
| "step": 2320 | |
| }, | |
| { | |
| "epoch": 0.7052972647125311, | |
| "grad_norm": 0.6353866457939148, | |
| "learning_rate": 0.00012240740740740742, | |
| "loss": 3.8061, | |
| "step": 2340 | |
| }, | |
| { | |
| "epoch": 0.7113254464622109, | |
| "grad_norm": 0.6012640595436096, | |
| "learning_rate": 0.0001187037037037037, | |
| "loss": 3.8055, | |
| "step": 2360 | |
| }, | |
| { | |
| "epoch": 0.7173536282118906, | |
| "grad_norm": 0.5284519195556641, | |
| "learning_rate": 0.000115, | |
| "loss": 3.7963, | |
| "step": 2380 | |
| }, | |
| { | |
| "epoch": 0.7233818099615703, | |
| "grad_norm": 0.45912453532218933, | |
| "learning_rate": 0.0001112962962962963, | |
| "loss": 3.8039, | |
| "step": 2400 | |
| }, | |
| { | |
| "epoch": 0.7233818099615703, | |
| "eval_loss": 3.754033095862669, | |
| "eval_perplexity": 42.69291988636338, | |
| "eval_runtime": 112.5914, | |
| "eval_samples_per_second": 14.735, | |
| "eval_steps_per_second": 0.231, | |
| "step": 2400 | |
| }, | |
| { | |
| "epoch": 0.7294099917112501, | |
| "grad_norm": 0.5166234970092773, | |
| "learning_rate": 0.00010759259259259259, | |
| "loss": 3.7997, | |
| "step": 2420 | |
| }, | |
| { | |
| "epoch": 0.7354381734609299, | |
| "grad_norm": 0.430136501789093, | |
| "learning_rate": 0.00010388888888888889, | |
| "loss": 3.7977, | |
| "step": 2440 | |
| }, | |
| { | |
| "epoch": 0.7414663552106096, | |
| "grad_norm": 0.5324243903160095, | |
| "learning_rate": 0.00010018518518518518, | |
| "loss": 3.7979, | |
| "step": 2460 | |
| }, | |
| { | |
| "epoch": 0.7474945369602893, | |
| "grad_norm": 0.39901381731033325, | |
| "learning_rate": 9.648148148148149e-05, | |
| "loss": 3.7905, | |
| "step": 2480 | |
| }, | |
| { | |
| "epoch": 0.7535227187099691, | |
| "grad_norm": 0.4086778163909912, | |
| "learning_rate": 9.277777777777778e-05, | |
| "loss": 3.7907, | |
| "step": 2500 | |
| }, | |
| { | |
| "epoch": 0.7595509004596489, | |
| "grad_norm": 0.45752742886543274, | |
| "learning_rate": 8.907407407407407e-05, | |
| "loss": 3.7895, | |
| "step": 2520 | |
| }, | |
| { | |
| "epoch": 0.7655790822093286, | |
| "grad_norm": 0.42594337463378906, | |
| "learning_rate": 8.537037037037038e-05, | |
| "loss": 3.7893, | |
| "step": 2540 | |
| }, | |
| { | |
| "epoch": 0.7716072639590084, | |
| "grad_norm": 0.42574238777160645, | |
| "learning_rate": 8.166666666666667e-05, | |
| "loss": 3.7837, | |
| "step": 2560 | |
| }, | |
| { | |
| "epoch": 0.7776354457086881, | |
| "grad_norm": 0.4005199074745178, | |
| "learning_rate": 7.796296296296296e-05, | |
| "loss": 3.7823, | |
| "step": 2580 | |
| }, | |
| { | |
| "epoch": 0.7836636274583678, | |
| "grad_norm": 0.4957292675971985, | |
| "learning_rate": 7.425925925925927e-05, | |
| "loss": 3.7763, | |
| "step": 2600 | |
| }, | |
| { | |
| "epoch": 0.7896918092080476, | |
| "grad_norm": 0.3952012360095978, | |
| "learning_rate": 7.055555555555556e-05, | |
| "loss": 3.782, | |
| "step": 2620 | |
| }, | |
| { | |
| "epoch": 0.7957199909577274, | |
| "grad_norm": 0.40845876932144165, | |
| "learning_rate": 6.685185185185186e-05, | |
| "loss": 3.7789, | |
| "step": 2640 | |
| }, | |
| { | |
| "epoch": 0.8017481727074072, | |
| "grad_norm": 0.3966081440448761, | |
| "learning_rate": 6.314814814814815e-05, | |
| "loss": 3.7788, | |
| "step": 2660 | |
| }, | |
| { | |
| "epoch": 0.8077763544570868, | |
| "grad_norm": 0.39544913172721863, | |
| "learning_rate": 5.9444444444444445e-05, | |
| "loss": 3.7706, | |
| "step": 2680 | |
| }, | |
| { | |
| "epoch": 0.8138045362067666, | |
| "grad_norm": 0.3594275712966919, | |
| "learning_rate": 5.5740740740740744e-05, | |
| "loss": 3.7731, | |
| "step": 2700 | |
| }, | |
| { | |
| "epoch": 0.8138045362067666, | |
| "eval_loss": 3.7262316295722786, | |
| "eval_perplexity": 41.52234140848831, | |
| "eval_runtime": 112.5682, | |
| "eval_samples_per_second": 14.738, | |
| "eval_steps_per_second": 0.231, | |
| "step": 2700 | |
| }, | |
| { | |
| "epoch": 0.8198327179564464, | |
| "grad_norm": 0.357453852891922, | |
| "learning_rate": 5.2037037037037035e-05, | |
| "loss": 3.771, | |
| "step": 2720 | |
| }, | |
| { | |
| "epoch": 0.8258608997061262, | |
| "grad_norm": 0.39895007014274597, | |
| "learning_rate": 4.8333333333333334e-05, | |
| "loss": 3.7732, | |
| "step": 2740 | |
| }, | |
| { | |
| "epoch": 0.8318890814558059, | |
| "grad_norm": 0.35610583424568176, | |
| "learning_rate": 4.462962962962963e-05, | |
| "loss": 3.7687, | |
| "step": 2760 | |
| }, | |
| { | |
| "epoch": 0.8379172632054857, | |
| "grad_norm": 0.40129202604293823, | |
| "learning_rate": 4.092592592592593e-05, | |
| "loss": 3.7664, | |
| "step": 2780 | |
| }, | |
| { | |
| "epoch": 0.8439454449551654, | |
| "grad_norm": 0.3733636736869812, | |
| "learning_rate": 3.722222222222222e-05, | |
| "loss": 3.7686, | |
| "step": 2800 | |
| }, | |
| { | |
| "epoch": 0.8499736267048451, | |
| "grad_norm": 0.32267412543296814, | |
| "learning_rate": 3.351851851851852e-05, | |
| "loss": 3.7709, | |
| "step": 2820 | |
| }, | |
| { | |
| "epoch": 0.8560018084545249, | |
| "grad_norm": 0.2760167717933655, | |
| "learning_rate": 2.9814814814814815e-05, | |
| "loss": 3.7632, | |
| "step": 2840 | |
| }, | |
| { | |
| "epoch": 0.8620299902042047, | |
| "grad_norm": 0.31765156984329224, | |
| "learning_rate": 2.6111111111111114e-05, | |
| "loss": 3.76, | |
| "step": 2860 | |
| }, | |
| { | |
| "epoch": 0.8680581719538845, | |
| "grad_norm": 0.30493825674057007, | |
| "learning_rate": 2.240740740740741e-05, | |
| "loss": 3.7644, | |
| "step": 2880 | |
| }, | |
| { | |
| "epoch": 0.8740863537035641, | |
| "grad_norm": 0.2905971109867096, | |
| "learning_rate": 1.8703703703703707e-05, | |
| "loss": 3.759, | |
| "step": 2900 | |
| }, | |
| { | |
| "epoch": 0.8801145354532439, | |
| "grad_norm": 0.2805236577987671, | |
| "learning_rate": 1.5e-05, | |
| "loss": 3.7634, | |
| "step": 2920 | |
| }, | |
| { | |
| "epoch": 0.8861427172029237, | |
| "grad_norm": 0.26742103695869446, | |
| "learning_rate": 1.1296296296296295e-05, | |
| "loss": 3.7545, | |
| "step": 2940 | |
| }, | |
| { | |
| "epoch": 0.8921708989526034, | |
| "grad_norm": 0.25122684240341187, | |
| "learning_rate": 7.592592592592593e-06, | |
| "loss": 3.751, | |
| "step": 2960 | |
| }, | |
| { | |
| "epoch": 0.8981990807022832, | |
| "grad_norm": 0.23693408071994781, | |
| "learning_rate": 3.888888888888889e-06, | |
| "loss": 3.7583, | |
| "step": 2980 | |
| }, | |
| { | |
| "epoch": 0.9042272624519629, | |
| "grad_norm": 0.24028028547763824, | |
| "learning_rate": 1.8518518518518518e-07, | |
| "loss": 3.7618, | |
| "step": 3000 | |
| }, | |
| { | |
| "epoch": 0.9042272624519629, | |
| "eval_loss": 3.711540102433069, | |
| "eval_perplexity": 40.91677404959386, | |
| "eval_runtime": 112.3619, | |
| "eval_samples_per_second": 14.765, | |
| "eval_steps_per_second": 0.231, | |
| "step": 3000 | |
| } | |
| ], | |
| "logging_steps": 20, | |
| "max_steps": 3000, | |
| "num_input_tokens_seen": 0, | |
| "num_train_epochs": 1, | |
| "save_steps": 300, | |
| "stateful_callbacks": { | |
| "TrainerControl": { | |
| "args": { | |
| "should_epoch_stop": false, | |
| "should_evaluate": false, | |
| "should_log": false, | |
| "should_save": true, | |
| "should_training_stop": true | |
| }, | |
| "attributes": {} | |
| } | |
| }, | |
| "total_flos": 1.253985226850304e+18, | |
| "train_batch_size": 64, | |
| "trial_name": null, | |
| "trial_params": null | |
| } | |